You are merging an external "Universal Agent Kit" (in ./universal-agent-kit/) into THIS
repository. My existing setup is the source of truth; the kit is an ADDITIVE library, not a
drop-in replacement. Your job is to integrate the useful parts without breaking my project.

INPUTS
- Kit: ./universal-agent-kit/
    CLAUDE.md                      project-rules template
    .claude/commands/*             slash-command skills (/spec, /spec-tech, /spec-dev,
                                   /spec-check, /spec-fix, /spec-all, /research, /quick, /fix,
                                   /git, /verify, /ui-clarify, /review, /caveman*)
    .claude/agents/*               role briefs (rd-lead, solution-researcher, implementer,
                                   doc-writer)
    .claude/settings.json          default agent + permission template
    docs/SPEC_LIFECYCLE.md         the ticket methodology
    docs/CODE_QUALITY.md           anti-slop conventions
    docs/AGENT_MEMORY.md           persistent memory discipline (four entry types)
    docs/RESEARCH_INDEX.md         research order + the queryable code index
    docs/VALIDATION.md             the validation ladder
    memory/                        memory index template + one sample entry per type
    README.md                      import guide + manifest
- Mine: my existing CLAUDE.md, .claude/, settings, conventions, and codebase.

HARD SAFETY RULES (non-negotiable)
1. Do NOT modify anything until I approve the plan (Step 2). Plan first, write second.
2. Do NOT overwrite or delete any of my existing files. On any name collision, keep mine and
   surface the conflict - never silently replace.
3. Do NOT touch ./universal-agent-kit/ itself (it is the read-only source).
4. Create a new git branch before any write. Do NOT commit, push, or rebase.
5. Back up any of my files you are about to change into a git-ignored scratch directory first
   (e.g. ./tmp/kit-merge-backup/) - never the repo root.
6. Do NOT loosen my permissions/settings. Propose additions for my review only.
7. Keep my conventions, naming, language, and architecture rules authoritative. Where the kit
   conflicts with mine, mine wins; mention the conflict so I can decide.

STEP 1 - INVENTORY (read-only)
- List what the kit offers vs what I already have: commands, agents, CLAUDE.md rules, docs,
  the memory system.
- Classify each kit item as: NEW (no equivalent of mine), DUPLICATE (I already have one), or
  CONFLICT (mine exists and differs in intent).
- Detect which agent tool THIS repo targets (Claude Code, Cursor, Cline, Windsurf, Aider, Codex).
  Only Claude Code loads .claude/commands/* and .claude/agents/* natively. For any other tool,
  plan to TRANSLATE each command/agent into that tool's rules format (mapping in
  universal-agent-kit/README.md) - do not copy .claude/ verbatim, the files would be dead.
- Grep the kit for <PLACEHOLDER> tokens; for each, propose a concrete value derived from THIS
  repo. The full set: <PROJECT_NAME>, <CHAT_LANGUAGE>, <INDEX_DOC>, <SRC_ROOT>, <ARCH_LAYERS>,
  <BUILD_CMD>, <TEST_CMD>, <LINT_CMD>, <RUN_CMD>, <LOGGER>, <PLAN_DIR>, <SCRATCH_DIR>, <MAX_LOC>,
  <READONLY_ZONES>, <ID>. Mark any you cannot resolve.
- If I have no persistent-memory setup yet, note whether my runtime supports one and how the
  kit's memory/ templates would map onto it.

STEP 2 - PLAN (stop and wait for my approval)
Present a concrete merge plan, then STOP:
- NEW items -> which files you'll copy in, to which paths.
- DUPLICATE/CONFLICT items -> for each, recommend one: skip, or import under a suffixed name
  (e.g. /spec-kit, agent "rd-lead-kit") so nothing of mine is shadowed. Show a short diff of
  mine vs the kit's so I can choose.
- CLAUDE.md -> show the exact rule blocks you propose to append/merge into MY CLAUDE.md (with
  placeholders filled). Never replace my CLAUDE.md wholesale.
- docs/ and memory/ -> which files you'll copy and where, and which CLAUDE.md line references
  each.
- settings.json -> show only the permission lines you'd suggest adding, as a proposal. Do not
  apply.
- The resolved placeholder table, plus any unresolved ones you need me to answer.
Wait for my "go" (or my edits) before writing anything.

STEP 3 - APPLY (only after I approve)
- On the new branch, back up touched files, then apply exactly the approved plan.
- Fill all approved placeholders. Leave unresolved ones as TODO and list them.
- Do not commit. End with: the file tree of what changed, the backup location, the unresolved
  TODOs, and how to try one new command safely.

Start with Step 1 now. Ask me questions only if something in Step 2 is genuinely undecidable
from the repo.
